COS 318: Internetworking

Or, how the Internet works Slides borrowed from Jennifer Rexford

The Internet according to Senator Ted Stevens

The Internet is not something you just dump something on. It's not a truck. It's a series of tubes. And if you don't understand, those tubes can be filled. And if they are filled, when you put your message in, it gets in line and it's going to be delayed by anyone that puts into that tube enormous amounts of material, enormous amounts of material.

No Truck, Yes Tubes

What the heck is going on in the Senate?

The Internet according to Wikipedia…

The Internet is the worldwide, publicly accessible network of interconnected computer networks that transmit data by packet switching using the standard Internet Protocol (IP). It is a "network of networks" that consists of millions of smaller domestic, academic, business, and government networks, which together carry various information and services, such as electronic mail, online chat, file transfer, and the interlinked Web pages and other documents of the World Wide Web.

http://en.wikipedia.org/wiki/Internet

Key Ideas Underlying the Internet

Idea #1: The rise of the stupid network

Telephone Network

Smart Network

Dumb Terminals

Telephone Network

u  Dumb phones l  Dial a number l  Speak and listen

u  Smart switches l  Set up and tear down a circuit l  Forward audio along the path

u  Limited services l  Audio l  Later, fax, caller-id, …

u  A monopoly for a long time

Internet

Dumb Network

Smart Terminals

Power at the Edge

End-to-End Principle Whenever possible, communications protocol operations should be defined to occur at the end-points of a communications system.

Programmability With programmable end hosts, new network services can be added at any time, by anyone.

And then end hosts became powerful and ubiquitous….

Let routers handle reliability /survivability? u  Need to replicate state u  End-point handling doesn’t require that u  Place more trust in end hosts u  Communication involving an end point doesn’t survive if

end-point goes down

Idea #2: Going Postal

Internet Protocol (IP) Packet Switching

u  Much like the postal system l  Divide information into letters l  Stick them in envelopes l  Deliver them independently l  And sometimes they get there

•  What’s in an IP packet? –  The data you want to send –  A header with the “from” and “to”

addresses

Why Packets? u  Data traffic is bursty

l  Logging in to remote machines l  Exchanging e-mail messages

u  Don’t waste bandwidth l  No traffic exchanged during idle periods

u  Better to allow multiplexing l  Different transfers share access to same links

tube

Why Packets? u  Packets can be delivered by most anything

l  Serial link, fiber optic link, coaxial cable, wireless u  Even birds

l  RFC 1149: IP Datagrams over Avian Carriers

IP over Avian Carriers was actually implemented, sending 9 packets over a distance of approximately 5km (3 miles), each carried by an individual pigeon, and they received 4 responses, with a packet loss ratio of 55%, and a response time ranging from 3000 seconds to over 6000 seconds.

Idea #3: Never having to say you’re sorry

Best-Effort Packet-Delivery Service

u Best-effort delivery l Packets may be lost l Packets may be corrupted l Packets may be delivered out of order

source destination

IP network

IP Service Model: Why Best-Effort? u  I’ve never promised you a rose garden

l  No error detection and correction l  Don’t remember from one packet to next l  Don’t reserve bandwidth and memory

u  Easier to survive failures l  Transient disruptions are okay during failover

u … but, applications do want efficient, accurate transfer of data in order, in a timely fashion

u  Let the end host take care of that!

Retransmit Lost and Delayed Packets

Internet GET index.html

Problem: Lost, Corrupted, or Delayed Data

Internet GET index.html

Solution: Timeout and Retransmit

GET index.html GET index.html

Waiting for an acknowledgment…

Discard Corrupted Packets

u  Sender computes a checksum l  Sender sums up all of the bytes l  And sends the sum to the receive

u  Receiver checks the checksum l  Received sums up all of the bytes l  And compares against the checksum

Internet GET index.html GET indey.html

134 + 212 = 346

134 + 216 = 350

Solution: Add Sequence Numbers

Problem: Out of Order

Putting Out of Order Packets Back in Order

GET x.ht inde ml

GET x.htindeml

GET index.html

ml 4 inde 2 x.ht 3 GET 1

Preventing Buffer Overflow at the Receiver u  Window size

l  Amount that can be sent without acknowledgment l  Receiver needs to be able to store this much data

u  Receiver advertises the window to sender l  Tells the receiver the amount of free space left l  … and sender agrees not to exceed this amount

Window Size

Outstanding Un-ack’d data

Data OK to send

Data not OK to send yet

Data ACK’d

Transmission Control Protocol (TCP)

u Communication service (socket) l  Ordered, reliable byte stream l  Simultaneous transmission in both directions

u Key mechanisms at end hosts l  Retransmit lost and corrupted packets l  Discard duplicate packets and put packets in order l  Flow control to avoid overloading the receiver buffer

source network destination

TCP connection

But, what if too many hosts send at once?

Idea #4: Think globally, act locally

Congestion

u  Too many hosts sending packets at once l  Some packets have to wait in line l  Eventually the queue runs out of space l  And some packets gets dropped on the floor

Sharing the Limited Resource u  Reserve resources

l  Room for ten phone calls l  Block the 11th call

u  Sub-divide resources l  Tell the 11 transfers to each use 1/11

of the bandwidth l  How????

u  Local adaptation l  Each transfer slows down l  Voluntarily, for greater good

Congestion Control

u  What if too many folks are sending data? l  Senders agree to slow down their sending rates l  … in response to their packets getting dropped l  For the greater good

TCP Congestion Control u  Detecting congestion

l  My packet was lost u  Reacting to congestion

l  I voluntarily reduce my sending rate (by 2X) u  Testing the waters

l  I gradually increase my sending rate (linearly)

send

ing

rate

Transmission Control Protocol (TCP) u  Runs on the end host

l  Puts data into packets and sends them u  Congestion control

l  Speeds up and slows down u  Ordered reliable byte stream

l  Sender retransmits lost packets l  Receiver discards corrupted packets l  Receiver reorders out-of-order packets

Reliable service on an unreliable network

Why not TCP for Everything? u  Applications have different needs

l  Latency, bandwidth, reliability u  E.g. real-time speech or video cares more about timing

than about getting all bytes u  So other transport protocols necessary u  Led to decoupling of TCP and IP

l  TCP: reliable ordered delivery l  IP: basic datagram service, best-effort delivery l  (UDP: application-level interface to basic datagram service)

Operating atop a variety of networks u  Internet operates over a variety of networks

l  Long-haul (X.25) l  Local-area (ethernet, token rings) l  Satellite l  Packet radio l  Serial links

u  Key: makes very few assumptions about underlying network capabilities l  Can transfer a packet l  Can address (unless point-to

Key idea #5: Layering

Layering: A Modular Approach u  Sub-divide the problem

l  Each layer relies on services from layer below l  Each layer exports services to layer above

u  Interface between layers defines interaction l  Hides implementation details l  Layers can change without disturbing other layers

Link hardware

Host-to-host connectivity

Application-to-application channels

Application

Application-Layer Protocols

u Messages exchanged between applications l  Syntax and semantics of the messages between hosts l  Tailored to the specific application (e.g., Web, e-mail) l  Messages transferred over transport connection (e.g.,

TCP)

u Popular application-layer protocols l  Telnet, FTP, SMTP, NNTP, HTTP, BitTorrent, …

Client Server GET /index.html HTTP/1.1

HTTP/1.1 200 OK

Layering in the Internet

HTTP

TCP

IP

Ethernet interface

HTTP

TCP

IP

Ethernet interface

IP IP

Ethernet interface

Ethernet interface

SONET interface

SONET interface

host host

router router

HTTP message

TCP segment

IP packet IP packet IP packet

Packet Encapsulation

Get index.html

Connection ID

Source/Destination

Link Address

User A User B

Packet Demultiplexing u  Multiple choices at each layer

FTP HTTP TFTP NV

TCP UDP

IP

NET1 NET2 NETn …

TCP/UDP IP

Port Number

Network

Protocol Field

Type Field

UDP TCP

Data Link

Physical

Applications

The Hourglass Model

Waist

The “narrow waist” facilitates interoperability

FTP HTTP TFTP NV

TCP UDP

IP

NET1 NET2 NETn …

The Narrow Waist of IP

Idea #6: A rose by any other name

Separating Naming and Addressing

u  Host names l  Mnemonic name appreciated by humans l  Variable length, alpha-numeric characters l  Provide little (if any) information about location l  Examples: www.cnn.com and ftp.eurocom.fr

u  IP addresses l  Numerical address appreciated by routers l  Fixed length, binary number l  Hierarchical, related to host location l  Examples: 64.236.16.20 and 193.30.227.161

Separating Naming and Addressing

u  Names are easier to remember l  www.cnn.com vs. 64.236.16.20

u  Addresses can change underneath l  www.cnn.com needn’t be at 64.236.16.20

u  Name could map to multiple IP addresses l  www.cnn.com to multiple replicas of the Web site

u  Map to different addresses in different places l  Address of a nearby copy of the Web site l  E.g., to reduce latency, or return different content

u  Multiple names for the same address l  E.g., aliases like ee.mit.edu and cs.mit.edu

Domain Name System (DNS) Hierarchy

u  Distributed “phone book” l  Multiple queries to translate name to address

u  Small number of “root servers” l  Tell you where to look up “.com” names

u  Larger number of “top-level domains” l  Tell you where to look up “cnn.com” names

root

.com

.edu

cnn.com

fox.com

DNS Resolver and Local DNS Server

Application

DNS resolver

Local DNS server

1 10

DNS cache

DNS query 2

DNS response 9

Root server

3

4

Top-level domain server

5

6

Second-level domain server

7

8

Caching to reduce latency in DNS translation.

Example: Many Steps in Web Download

Browser cache

DNS resolution

TCP open

1st byte response

Last byte response

Sources of variability of delay •  Browser cache hit/miss, need for cache revalidation •  DNS cache hit/miss, multiple DNS servers, errors •  Packet loss, round-trip time, server accept queue •  RTT, busy server, CPU overhead (e.g., CGI script) •  Response size, receive buffer size, congestion •  … downloading embedded image(s) on the page

Idea #7: You scratch my back…

Network of Networks

1

2

3

4

5

6 7

Autonomous Systems

Autonomous Systems

•  Level 3: #1 •  MIT: #3 •  Harvard: #11 •  Yale: #29 •  Princeton: #88 •  AT&T: #7018, #6341, #5074, … •  UUNET: #701, #702, #284, #12199, … •  Sprint: #1239, #1240, #6211, #6242, … •  …

Currently around 20,000 ASes.

Cooperation and Competition

1

2

3

4

5

6 7

ClientWeb server

Traffic flows through many ASes

Interdomain routing protocol

Business Relationships

u  Neighboring ASes have business contracts l  How much traffic to carry l  Which destinations to reach l  How much money to pay

u  Common business relationships l  Customer-provider

•  E.g., Princeton is a customer of AT&T and USLEC •  E.g., MIT is a customer of Level3

l  Peer-peer •  E.g., AT&T is a peer of Sprint •  E.g., Harvard is a peer of Harvard Business School

Problems With the Internet: Cheaters do win

No Strict Notions of Identity

u  Leads to l  Spam l  Spoofing l  Denial-of-service

Nobody in Charge u  Traffic traverses many Autonomous Systems

l  Who’s fault is it when things go wrong? l  How do you upgrade functionality?

u  Implicit trust in the end host l  What if some hosts violate congestion control?

u  Anyone can add any application l  Whether or not it is legal, moral, good, etc.

u  Nobody knows how big the Internet is l  No global registry of the topology

u  Spans many countries l  So no government can be in charge

The Internet of the Future

u  Can we fix what ails the Internet l  Security l  Performance l  Upgradability l  Managability l  <your favorite gripe here>

u  Without throwing out the baby with bathwater l  Ease of adding new hosts l  Ease of adding new services l  Ease of adding new link technologies

u  An open technical and policy question…